Inspection device, injection molding system, and inspection method

Abstract

The present application relates to an inspection device capable of determining the quality of a molded product with a small load, an injection molding system, and an inspection method. The inspection device includes an imaging device that acquires a polarization image of a molded product, and a determination unit that determines the quality of the molded product. The determination unit has an image generation unit that generates a pseudo image from an input image according to a learned learning model, and inputs the polarization image or a calculation image obtained by calculating from the polarization image to the image generation unit as the input image, and determines the quality of the molded product according to the input image and the pseudo image generated by the image generation unit.

Description

Technical Field

[0001] This application claims priority based on Japanese Patent Application No. 2022-056149, filed on March 30, 2022. The entire contents of that Japanese application are incorporated herein by reference.

[0002] This invention relates to inspection devices, injection molding systems, and inspection methods. Background Technology

[0003] Patent document 1 discloses a system comprising: a light source that illuminates an injection-molded article through a polarizer; a camera component that photographs the injection-molded article through the polarizer; and a setting value correction component that corrects the setting value of the molding step based on the polarized light stripe pattern obtained by the camera component.

[0004] Patent Document 1: Japanese Patent Application Publication No. 1-120317 Summary of the Invention

[0005] Preferably, the quality of injection-molded articles can be determined with a relatively small load. The object of the present invention is to provide an inspection device, injection molding system, and inspection method that can determine the quality of molded articles with a relatively small load.

[0006] The inspection device involved in this invention includes:

[0007] A camera device is used to acquire polarization images of the molded product; and

[0008] The determination unit determines the quality of the molded article.

[0009] The determination unit includes an image generation unit that generates pseudo-images from input images based on a machine learning model.

[0010] The determination unit inputs the polarization image or the processed image obtained from the polarization image as the input image to the image generation unit, and determines the quality of the molded product based on the input image and the pseudo image generated by the image generation unit.

[0011] The injection molding system of the present invention includes an injection molding machine and the aforementioned inspection device for inspecting the molded articles formed by the injection molding machine.

[0012] The inspection method involved in this invention is as follows:

[0013] A polarization image of the molded product is acquired using a camera device;

[0014] The polarization image or the processed image obtained from the polarization image is input as an input image to the image generation unit; and

[0015] The quality of the molded product is determined based on the pseudo image of the input image generated by the image generation unit and the input image itself.

[0016] The effects of the invention

[0017] According to the present invention, an inspection device, an injection molding system, and an inspection method are provided that can determine the quality of molded articles with a relatively small load. Attached Figure Description

[0018] Figure 1 This is a diagram illustrating an example of the schematic structure of the injection molding system involved in the embodiment.

[0019] Figure 2 This is an example of a block diagram illustrating the functions of a control device and a processing device.

[0020] Figure 3 This is a diagram illustrating an example of the general structure of a camera device.

[0021] Figure 4 This is a diagram illustrating an example of a polarized image acquired by a camera device.

[0022] Figure 5 This is a diagram illustrating a schematic structure of the camera device in Modified Example 1.

[0023] Figure 6 This is a diagram illustrating a schematic structure of the camera device in Modified Example 2.

[0024] Figure 7 This is a diagram illustrating a schematic structure of the camera device in Modified Example 3.

[0025] Figure 8 This is a flowchart illustrating an example of the sequence of inspection processes performed by the processing device.

[0026] Figure 9 This is a diagram illustrating the inspection process of the first embodiment.

[0027] Figure 10 This is a diagram illustrating the inspection process of the second embodiment.

[0028] Figure 11 This is a diagram illustrating the inspection process of the third embodiment.

[0029] Figure 12 This is a diagram illustrating the inspection process of the fourth embodiment.

[0030] Figure 13 This is a diagram illustrating the inspection process of the fifth embodiment.

[0031] Figure 14 This is an image representing an example of the inspection results.

[0032] Figure 15 This is a flowchart illustrating an example of the sequence of machine learning processes performed by the inspection device.

[0033] Figure 16 (A) is a diagram illustrating the machine learning processing of embodiments 1 to 4. Figure 16 (B) is a diagram illustrating the machine learning process of the fifth embodiment.

[0034] Explanation of symbols

[0035] 1-Injection molding system, 2-Injection molding machine, 3-Molded product, 10-Injection device, 20-Screw, 40-Control device, 60-Drive device, 61-Metering motor, 71-Injection motor, 100-Inspection device, 110-Camera device, 111-Light source, 112-Linear polarizer, 113-Wavelength plate, 114-Polarized light camera, 120-Processing device, 131-Receiving unit, 132-Judgment unit, 132a-Image generation unit, 132b-First image generation unit, 132c-Second image generation unit, 133-Output unit, 215, 315-Rotating components, 141-Operating unit, 142-Training data input unit, 143-CAE calculation unit, 144-Molding step data input unit, pin-Input image, pout-Pseudo-image, e1-Judgment result. Detailed Implementation

[0036] Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings.

[0037] Figure 1 This is a diagram showing an example of the schematic structure of the injection molding system 1 according to the first embodiment. Figure 2 This is an example of a block diagram illustrating the functions of the control device 40 and the processing device 120.

[0038] The injection molding system 1 includes an injection molding machine 2 and an inspection device 100 for checking the quality (whether it is normal or abnormal) of the molded articles produced by the injection molding machine 2.

[0039] Injection Molding Machine 2

[0040] First, the injection molding machine 2 will be described. In the following description, the direction of resin injection is defined as the front side, and the direction opposite to the direction of resin injection is defined as the rear side.

[0041] The injection molding machine 2 includes a mold clamping device (not shown), an injection device 10, a material supply device 81, a control device 40 for the entire control device, an operation unit 51 for receiving user input operations, and a display unit 52 for displaying operation receiving screens and images.

[0042] The mold clamping device, injection device 10, material supply device 81, and control device 40 will be described in detail later.

[0043] The operation unit 51 can be exemplified as an input device such as a button, switch, or touch panel. The display unit 52 can be exemplified as a liquid crystal display or an organic EL display. The operation unit 51 and the display unit 52 can also be integrated into one unit.

[0044] The injection molding machine 2 repeatedly manufactures molded products in one cycle, consisting of mold closing, mold clamping, filling, holding pressure, cooling, metering, mold opening, and ejection. The mold closing process involves closing the mold assembly, which consists of a fixed mold and a movable mold. The mold clamping process secures the mold assembly. The filling process involves allowing molten resin to flow into the mold assembly. The holding pressure process applies pressure to the flowing resin. The cooling process solidifies the resin within the mold assembly after the holding pressure process. The metering process measures the molten resin used for the next molded product. The mold opening process opens the mold assembly. The ejection process ejects the molded product from the mold assembly after mold opening. Additionally, to shorten the molding cycle, the metering process can be performed during the cooling process.

[0045] (Mold closing device)

[0046] The mold closing device includes a fixed pressure plate for mounting a fixed mold and a movable pressure plate for mounting a movable mold. By moving the movable pressure plate forward and backward, the movable mold is separated from the fixed mold to perform mold closing, mold assembly, and mold opening. There are no particular limitations on the type of mold closing device. Examples include a toggle type using an electric motor and a toggle mechanism, a direct-pressure type using a fluid pressure cylinder, and an electromagnetic type using a linear motor and an electromagnet.

[0047] (Injection device 10)

[0048] The injection device 10 includes: a cylinder 11 for heating resin as a molding material; and a nozzle 12 disposed at the front end of the cylinder 11. Furthermore, the injection device 10 includes: a screw 20 rotatable within the cylinder 11 and freely movable in the direction of its rotation axis; heaters h11, h12, and h13 serving as a heat source for heating the cylinder 11; and a drive device 60 disposed at the rear side of the cylinder 11.

[0049] The screw 20 has a screw body 21 and an injection section 22 disposed further forward than the screw body 21, and is connected to the drive device 60 via a shaft at the rear end. The screw body 21 has a threaded portion 23 and a pressure member 24 detachably disposed relative to the front end of the threaded portion 23. The threaded portion 23 has a rod-shaped body portion 23a and a helical thread 23b formed in a manner protruding from the outer peripheral surface of the body portion 23a, and a helical thread groove 26 is formed along the thread 23b. It can be exemplified that the depth of the thread groove 26 is constant from the rear end to the front end of the threaded portion 23, and the screw compression ratio is constant.

[0050] Alternatively, the screw 20 may not have a pressure member 24, but instead have a threaded portion 23 formed on the entire screw body 21. Furthermore, the screw body 21 may be divided from the rear end to the front end into a supply section for supplying resin, a compression section for melting the supplied resin while compressing it, and a metering section for metering a fixed amount of molten resin each time. Preferably, the depth of the threaded groove 26 is deepest in the supply section, shallowest in the metering section, and becomes shallower in the compression section as it moves from the rear to the front.

[0051] The injection part 22 has: a head 31 with a conical portion at the front end; a rod 32 formed adjacent to the rear side of the head 31; a check ring 33 disposed around the rod 32; and a sealing ring 34 installed at the front end of the pressure member 24.

[0052] During the metering process, as the screw 20 retracts, the check ring 33 moves forward relative to the rod 32. When it separates from the sealing ring 34, resin is conveyed from the rear to the front of the injection section 22. Furthermore, during the injection process, as the screw 20 advances, the check ring 33 moves rearward relative to the rod 32. When it contacts the sealing ring 34, it prevents backflow of resin.

[0053] A resin supply port 14, serving as a molding material supply port, is formed at the rear of the cylinder body 11. The resin supply port 14 is formed at a position opposite to the rear end of the threaded groove 26 when the screw 20 is positioned at the foremost side inside the cylinder body 11. A material supply device 81 for supplying resin into the cylinder body 11 is mounted on the resin supply port 14.

[0054] The drive unit 60 is a device that rotates or retracts the screw 20 within the cylinder 11. The drive unit 60 includes a metering motor 61 as a drive source for rotating the screw 20 within the cylinder 11 and an injection motor 71 as a drive source for moving the screw 20 within the cylinder 11 along the rotation axis. The metering motor 61 and the injection motor 71 can be exemplified as servo motors.

[0055] A motion conversion mechanism is provided between the injection motor 71 and the screw 20 to convert the rotational motion of the injection motor 71 into the linear motion of the screw 20. For example, the motion conversion mechanism has a lead screw shaft and a lead screw nut screwed to the lead screw shaft. Ball bearings, rollers, etc., can be provided between the lead screw shaft and the lead screw nut. The drive source for moving the screw 20 along the rotation axis is not limited to the injection motor 71; for example, it can be a hydraulic cylinder, etc.

[0056] (Material supply device 81)

[0057] The material supply device 81 includes: a hopper 82 for holding molding material (e.g., resin granules); a feed cylinder 83 extending horizontally from the lower end of the hopper 82; and a cylindrical guide 84 extending downward from the front end of the feed cylinder 83. Furthermore, the material supply device 81 includes: a feed screw 85 rotatably disposed within the feed cylinder 83; and a feed motor 86 for rotating the feed screw 85.

[0058] The resin supplied from the hopper 82 to the feed cylinder 83 advances along the threaded groove of the feed screw 85 as it rotates. The resin conveyed from the front end of the feed screw 85 to the guide section 84 falls into the guide section 84 and is then supplied to the cylinder 11.

[0059] Furthermore, the feed cylinder 83 does not necessarily have to extend horizontally; for example, it can extend at an angle relative to the horizontal direction. Also, the outlet side of the feed cylinder 83 can be higher than the inlet side.

[0060] Furthermore, the resin supplied to the feed cylinder 83 can also be heated by a heater (not shown). In this case, it is preferable to heat the resin to a temperature at which it will not melt, such as a predetermined temperature below the glass transition point.

[0061] (Control device 40)

[0062] The control device 40 includes a CPU 41, a ROM 42 for storing control programs, a read / write RAM 43 for storing calculation results, a storage unit 44 such as a hard disk, an input / output interface (I / F) 45, and an output / output interface (I / F) 46. The control device 40 performs various functions by having the CPU 41 execute programs stored in the ROM 42 or the storage unit 44.

[0063] The control device 40 may include: a motor control unit 47, which controls the drive of the metering motor 61, the injection motor 71, the feed motor 86, etc.; a heater control unit 48, which controls the temperature of the heaters h11 to h13; and a parameter correction unit 49, which corrects the molding parameters when molding the molded article 3.

[0064] (Operation of injection molding machine 2)

[0065] The operation of the injection molding machine 2 controlled by the control device 40 will now be described. During the metering process, the motor control unit 47 of the control device 40 drives the metering motor 61 to rotate the screw 20. At the same time, the motor control unit 47 drives the feed motor 86 to rotate the feed screw 85. It can be illustrated that the motor control unit 47 rotates the screw 20 and the feed screw 85 synchronously during molding. The motor control unit 47 controls the current supplied to the metering motor 61 so that the rotational speed of the screw 20 is, for example, a speed set by the operation unit 51. Furthermore, the motor control unit 47 controls the current supplied to the feed motor 86 so that the rotational speed of the feed screw 85 is, for example, a speed set by the operation unit 51.

[0066] The resin supplied to the cylinder 11 by the material supply device 81 does not stagnate at the resin supply port 14, but is directly conveyed to the front by the screw 20. The resin does not tightly fill the threaded groove 26 of the screw 20; the resin in the threaded groove 26 is in a sparse state. Therefore, the faster the resin is supplied by the material supply device 81, the greater the amount of resin conveyed to the front by the screw 20 per unit time.

[0067] The resin supplied to the cylinder 11 moves along the threaded groove 26 of the screw 20 as the screw 20 rotates, and is heated and melted by the heaters h11 to h13. The heater control unit 48 of the control device 40 controls the power supplied to the heaters h11 to h13 so that the temperature of the heaters h11 to h13 is, for example, the temperature set by the operation unit 51.

[0068] Furthermore, the resin supplied to the cylinder 11 is gradually pressurized from the pressure rise start position of the resin in the screw body 21 to the front end of the screw body 21. The pressure rise start position is located at a predetermined distance from the pressure member 24, and the displacement is based on the ratio (synchronization rate) of the rotational speed of the screw 20 to the rotational speed of the feed screw 85. When the pressure rise start position is within a predetermined distance from the pressure member 24, the molten state of the resin stabilizes, and the weight of the molded product stabilizes.

[0069] The resin advancing along the threaded grooves 26 of the screw 20 passes through the resin flow path between the pressure member 24 and the cylinder 11, is mixed during this period, and then advances through the resin flow path between the cylinder 11 and the rod portion 32. The resin is then conveyed to the front of the screw 20 and accumulates at the front of the cylinder. As molten resin accumulates at the front of the screw 20, the screw 20 retracts.

[0070] In the metering process, the motor control unit 47 of the control device 40 controls the current supplied to the injection motor 71 so that the back pressure of the screw 20 becomes, for example, the back pressure set by the operation unit 51. By applying back pressure to the screw 20, the rapid retraction of the screw 20 is suppressed, the composability of the resin is improved, and the gas in the resin can easily escape to the rear.

[0071] During the retraction of the screw 20, the motor control unit 47 monitors the position of the screw 20 using a position sensor (not shown). When the screw 20 retracts to the metering completion position and a predetermined amount of resin accumulates on the front side of the screw 20, the control device 40 stops driving the metering motor 61. Thus, the rotation of the screw 20 stops, and the metering process is completed. It can be exemplified that the motor control unit 47 stops driving the feed motor 86 and stops the rotation of the feed screw 85 simultaneously with the completion of the metering process.

[0072] During the filling process, the motor control unit 47 of the control device 40 drives the injection motor 71, causing the screw 20 to advance and push the resin into the cavity space within the mold device in the closed state. At this time, the motor control unit 47 controls the current supplied to the injection motor 71 so that the moving speed of the screw 20 in the rotational axis direction becomes, for example, the moving speed set by the operation unit 51.

[0073] During the pressure holding process, the motor control unit 47 controls the current supplied to the injection motor 71 so that the resin pressure becomes, for example, the pressure set by the operation unit 51. As a result, the resin filling the cavity space shrinks due to cooling, but the amount of shrinkage is replenished by the resin.

[0074] In addition, the set values ​​of rotation speed, movement speed, pressure, etc. used by the motor control unit 47 when controlling various motors, and the set values ​​of temperature used by the heater control unit 48 when controlling heaters h11 to h13 are stored as molding parameters in ROM 42 or storage unit 44, etc.

[0075] The processing unit 120 of the inspection device 100 sends a quality judgment of the molded product to the parameter correction unit 49. The parameter correction unit 49 uses the judgment result that the molded product 3 is a defective product output from the processing unit 120 and the molding parameters when the molded product 3 is molded in the injection molding machine 2 to correct the molding parameters when the molded product 3 is molded next time.

[0076] If the processing unit 120 outputs a result indicating that the molded article 3 is a defective product, the parameter correction unit 49 determines that the molding parameters used to mold the article 3 in the injection molding machine 2 are unsuitable. Examples of molding parameters include the screw 20's moving speed during the filling process and the temperatures of the heaters h11 to h13. For example, it can be considered that wear on the mold assembly used for the molded article 3 deteriorates the fluidity of the molten resin within the mold assembly, thus altering the stress distribution. Therefore, the parameter correction unit 49 can be used to change at least one of the screw 20's moving speed and the heaters h11 to h13's temperatures. This is because if the screw 20's moving speed increases, the fluidity of the molten resin within the mold assembly improves; if the temperatures of the heaters h11 to h13 increase and the temperature of the molten resin increases, the fluidity of the molten resin within the mold assembly also improves.

[0077] Inspection Device 100

[0078] like Figure 1 As shown, the inspection device 100 includes a camera device 110 for photographing the molded article 3 and a processing device 120 for processing the images output from the camera device 110.

[0079] (Camera device 110)

[0080] Figure 3 This is a diagram showing an example of the schematic structure of the camera device 110. Figure 4 This is a diagram showing an example of a polarized image acquired by the camera device 110.

[0081] The imaging device 110 takes the molded article 3 formed by the injection molding machine 2 as the subject of the photograph and acquires a polarized image of the molded article 3 including multiple polarization channels. A polarized image refers to the image of the subject that appears through the refraction and reflection of polarized light. Assuming that for the same subject, taken from the same angle, direction, and distance, different types of polarized light are used for imaging, different images can be obtained for the same subject. That is, by using multiple types of polarized light for imaging, multiple components of the image can be obtained for the same subject. A polarization channel refers to the image of one component among the multiple components of the image.

[0082] The camera device 110 includes: a light source 111 that generates light; a linear polarizer 112 that generates linearly polarized light from the light emitted from the light source 111; a wavelength plate 113 that converts the linearly polarized light generated by the linear polarizer 112 into circularly polarized light; and a polarized light camera 114.

[0083] Light source 111 can be exemplified as a light bulb, incandescent lamp, fluorescent lamp, LED, etc. The light generated from light source 111 is not limited to visible light, but can also be infrared light. Light with a wavelength of 360 to 900 nm is preferred.

[0084] The linear polarizer 112 is an optical element that generates linearly polarized light from light emitted from the light source 111.

[0085] Wavelength plate 113 can be exemplified as a 1 / 4 wavelength plate that generates a phase difference of 90 degrees.

[0086] The polarization camera 114 can be exemplified as a camera in which polarizers at 0 degrees, 45 degrees, 90 degrees, and 135 degrees are regularly arranged between the imaging element and the lens, enabling the acquisition of a polarization image including four polarization channels corresponding to the aforementioned four polarization angles in a single shot. The polarization camera 114 can also be a camera capable of generating an image by performing calculations (e.g., arithmetic operations, trigonometric function operations, and inverse trigonometric function operations) on the polarization image to determine the direction and degree of polarization of the polarized light.

[0087] In the imaging device 110 configured as described above, the molded article 3 formed by the injection molding machine 2 is placed above the wavelength plate 113, and the light transmitted through the molded article 3 is captured by the polarized light camera 114.

[0088] like Figure 4 As shown, the above-described shooting method can acquire polarization images of four polarization channels corresponding to four polarization angles.

[0089] Alternatively, in the imaging device 110, the wavelength plate 113 can be omitted, and linearly polarized light transmitted through the linear polarizer 112 can be incident on the molded article 3. However, in the case where the structure allows linearly polarized light to be incident on the molded article 3, when the principal axis direction of the molded article 3 is orthogonal to the polarization direction, the light cannot pass through the molded article 3, and thus information may be lost. Therefore, it is preferable to use the wavelength plate 113 to allow circularly polarized light to be incident on the molded article 3. By allowing circularly polarized light to be incident on the molded article 3, information loss can be suppressed.

[0090] Furthermore, the wavelength plate 113 can also be a λ / 2 plate. If a λ / 2 plate with its optical axis tilted at a 45-degree angle is used as the wavelength plate 113, the polarization direction can be rotated vertically. This allows the polarization direction of the linearly polarized light transmitted through the linear polarizer 112 to be changed before it is incident on the molded article 3, suppressing information loss. That is, when the principal axis of the molded article 3 is orthogonal to the polarization direction, and light cannot pass through the molded article 3, a fringe image corresponding to the desired stress distribution cannot be output. Therefore, by using a λ / 2 plate to vertically reverse the polarization direction, information loss caused by light not being able to pass through the molded article 3 can be suppressed.

[0091] Furthermore, the wavelength plate 113 can also be a λ / 8 plate that converts linearly polarized light into elliptically polarized light, etc.

[0092] (A variation of a polarized light camera)

[0093] Figure 5 This is a diagram illustrating a schematic structure of the camera device 110 in Modified Example 1. Figure 6 This is a diagram illustrating a schematic structure of the camera device 110 in Modified Example 2. Figure 7 This is a diagram illustrating a schematic structure of the camera device 110 in Modified Example 3. Next, modified examples of the camera device 110 will be described. Components identical to those in the camera device 110 described above will be represented by the same reference numerals, and detailed descriptions will be omitted.

[0094] (Variation Example 1)

[0095] like Figure 5 As shown, the imaging device 110 of Modified Example 1 includes a light source 111, a linear polarizer 112, a polarized light camera 114, and a rotating member 215 for rotating the linear polarizer 112. The rotating member 215 rotates the film-shaped linear polarizer 112 by 45 degrees, 90 degrees, and 135 degrees with a line orthogonal to the plate surface as the center of rotation.

[0096] In the imaging device 310 configured as described above, the molded article 3 formed by the injection molding machine 2 is positioned above the linear polarizer 112, and the light transmitted through the molded article 3 is captured by the polarizing camera 114. Furthermore, the linear polarizer 112 is rotated by the rotating member 215 at 45 degrees, 90 degrees, and 135 degrees, and the polarizing camera 114 captures the light transmitted through the molded article 3 at each rotation angle. The polarizing camera 114 acquires images corresponding to the four polarization angles at each rotation angle of 0 degrees, 45 degrees, 90 degrees, and 135 degrees, and uses these images to calculate the polarization direction / degree of polarization. The structure and rotation method of the rotating member 215 that rotates the linear polarizer 112 are not particularly limited. The linear polarizer 112 can be rotated by a robot or manually.

[0097] In the camera device 110 configured in this way, it is also possible to acquire polarization images of four polarization channels corresponding to four polarization angles. Figure 4 ).

[0098] (Variation Example 2)

[0099] like Figure 6 As shown, the imaging device 110 of Modified Example 2 includes a light source 111, a linear polarizer 112, a polarized light camera 114, and a rotating member 315 for rotating the molded article 3. The rotating member 315 rotates the molded article 3 by 45 degrees, 90 degrees, and 135 degrees with a line orthogonal to the plate surface of the film-shaped linear polarizer 112 as the rotation center.

[0100] In the imaging device 310 configured as described above, the molded article 3 formed by the injection molding machine 2 is positioned above the linear polarizer 112, and the light transmitted through the molded article 3 is captured by the polarizing camera 114. Furthermore, the molded article 3 is rotated by a rotating member 315 at 45 degrees, 90 degrees, and 135 degrees, and the polarizing camera 114 captures the light transmitted through the molded article 3 at each rotation angle. The polarizing camera 114 acquires images corresponding to the four polarization angles at each rotation angle of 0 degrees, 45 degrees, 90 degrees, and 135 degrees, and uses these images to calculate the polarization direction / degree of polarization. The structure and rotation method of the rotating member 315 that rotates the molded article 3 are not particularly limited. The molded article 3 can be rotated by a robot or by hand.

[0101] In the camera device 110 configured in this way, it is also possible to acquire polarization images of four polarization channels corresponding to four polarization angles. Figure 4 ).

[0102] (Variation Example 3)

[0103] like Figure 7 As shown, the imaging device 110 according to Modification 3 includes a light source 111, a linear polarizer 112, and a wavelength plate 113. Furthermore, the imaging device 110 includes: a first beam splitter 421 that reflects a portion of the light emitted from the light source 111 and transmits a portion therethrough; a second beam splitter 422 that reflects a portion of the light transmitted through the first beam splitter 421 and transmits a portion therethrough; and a third beam splitter 423 that reflects a portion of the light transmitted through the second beam splitter 422 and transmits a portion therethrough. The imaging device 110 also includes: a first linear polarizer 431 that generates linearly polarized light from the light reflected by the first beam splitter 421; and a first camera 441 that captures images of the light transmitted through the first linear polarizer 431. Furthermore, the imaging device 110 includes: a second linear polarizer 432 that generates linearly polarized light from light reflected by the second beam splitter 422; and a second camera 442 that captures images of the light transmitted through the second linear polarizer 432. The imaging device 110 also includes: a third linear polarizer 433 that generates linearly polarized light from light reflected by the third beam splitter 423; and a third camera 443 that captures images of the light transmitted through the third linear polarizer 433. Finally, the imaging device 110 includes: a fourth linear polarizer 434 that generates linearly polarized light from light transmitted through the third beam splitter 423; and a fourth camera 444 that captures images of the light transmitted through the fourth linear polarizer 434.

[0104] Cameras 441 to 444 have imaging elements and lenses, but unlike polarized camera 114, they are ordinary cameras without polarizers at 0 degrees, 45 degrees, 90 degrees, and 135 degrees.

[0105] The polarization axis (transmission axis) of the second linear polarizer 432 is tilted at 45 degrees relative to the polarization axis of the first linear polarizer 431. The polarization axis (transmission axis) of the third linear polarizer 433 is tilted at 90 degrees relative to the polarization axis of the first linear polarizer 431. The polarization axis (transmission axis) of the fourth linear polarizer 434 is tilted at 135 degrees relative to the polarization axis of the first linear polarizer 431.

[0106] In the imaging device 110 configured as described above, the molded article 3 formed by the injection molding machine 2 is placed above the wavelength plate 113, and the light transmitted through the molded article 3 is captured by the first camera 441 to the fourth camera 444.

[0107] In the camera device 110 configured in this way, it is also possible to acquire polarization images of four polarization channels corresponding to four polarization angles. Figure 4 ).

[0108] (Processing device 120)

[0109] like Figure 1 As shown, the processing device 120 includes a CPU 121, a ROM 122 for storing control programs, a read / write RAM 123 for storing calculation results, a storage unit 124 such as a hard disk, an input interface (I / F) 125, and an output interface (I / F) 126. The processing device 120 performs various functions by executing programs stored in the ROM 122 or the storage unit 124 by the CPU 121.

[0110] like Figure 2 As shown, the processing device 120 includes: a receiving unit 131 that receives a polarization image of the molded article 3 output from the imaging device 110; a determination unit 132 that uses the polarization image of the molded article 3 received by the receiving unit 131 to determine the quality of the molded article 3; and an output unit 133 that outputs the result determined by the determination unit 132 to the control device 40.

[0111] Figure 8 This is a flowchart illustrating an example of the sequence of inspection processes performed by the processing device 120.

[0112] The processing device 120 performs the processing repeatedly, for example, at a predetermined control cycle (e.g., every 1 second).

[0113] The processing device 120 determines whether the receiving unit 131 has received a polarization image of the molded product 3 from the imaging device 110 (S501). If no polarization image is received ("No" in S501), the processing device 120 ends the inspection process. On the other hand, if a polarization image is received ("Yes" in S501), the processing device 120 inputs the polarization image received in S501 or the processed image calculated from the polarization image to the image generation unit 132a to generate a pseudo image (S502). Furthermore, the processing device 120 calculates the difference between the polarization image or the processed image and the pseudo image (S503). Then, the processing device 120 determines whether the difference calculated in S503 is above a predetermined threshold (S504). Furthermore, if the difference is above the threshold ("Yes" in S504), the processing device 120 determines that the molded product 3 is abnormal (defective product) (S505). On the other hand, if the difference is less than the threshold ("No" in S504), the processing device 120 determines that the molded product 3 is normal (qualified) (S506). Then, the processing device 120 outputs the determination result of whether the molded product 3 is qualified or unqualified to the control device 40 (S507). The processing of steps S502, S503, S504, S505, and S506 is performed by the determination unit 132, and the processing of step S507 is performed by the output unit 133.

[0114] Judgment Section 132

[0115] Next, regarding the structure of the determination unit 132 and in Figure 8 The detailed contents of the inspection process performed in steps S502 to S507 will be described for the first to fifth embodiments. Figures 9-13 These are diagrams illustrating the determination units of the first to fifth embodiments.

[0116] (Determination Unit of the First Embodiment)

[0117] like Figure 9 As shown, the determination unit 132 includes an image generation unit 132a that generates a pseudo-image from the input image based on a machine learning model. The determination unit 132 also includes a processing unit 132p1 that calculates the difference between the pseudo-image and the input image; and a processing unit 132p2 that compares the difference with a threshold and determines the quality of the molded product 3 based on the input image, the processed image, and the pseudo-image.

[0118] The image generation unit 132a of the first embodiment inputs a polarization image with four polarization channels, obtained by photographing the molded part of the inspection object, as an input image. Furthermore, the image generation unit 132a generates a pseudo image with four polarization channels. Even if the molded part in the input image contains an anomaly, the image generation unit 132a generates a pseudo image possessing the characteristics of a normal molded part.

[0119] The image generation unit 132a includes a GAN (Generative Adversarial Network) as a learning model. A GAN is a learning model that includes a generator and a discriminator. This learning model is pre-loaded with images of multiple sample molded products without anomalies (i.e., normal) as training data, and machine learning is performed using these images. The multiple sample molded products may include a number of sample molded products with different states within the normal range. Through this machine learning, the image generation unit 132a generates pseudo-images that possess the characteristics of normal molded products.

[0120] If a pseudo-image is obtained, the determination unit 132 (processing unit 132p1) calculates the difference between the input image and the pseudo-image. Furthermore, the determination unit 132 (processing unit 132p2) determines that if the difference is below a threshold, the molded product is normal; if the difference exceeds the threshold, the molded product is abnormal.

[0121] In the output processing of the determination result (S507), the output unit 133 can output the input image, the pseudo image, and the determination result of whether it is normal or abnormal to the display unit 52. Moreover, the determination unit 132 can also display the difference between the input image and the pseudo image, as well as the areas with larger differences for each pixel.

[0122] Regarding the difference between the input image and the pseudo image, any quantity that can represent the difference can be used. For example, the following three quantities can be used.

[0123] Firstly, the aforementioned gap can be addressed by using the sum of differences between the input image and the pseudo-image. The sum of differences refers to the sum of the differences (absolute value, squared value, etc.) of the values ​​(brightness values) of the same pixels in the same polarization channel for all polarization channels and all pixels.

[0124] Secondly, the aforementioned gap can be calculated using the value of a certain variable (values ​​from multiple groups) in the latent variable space of the GAN of the image generation unit 132a. As a certain variable in the latent variable space, a variable reflecting the distance between the input image and the pseudo-image can be found in the latent variable space and used. When using the value of the aforementioned variable, the sum of the distance calculated from the aforementioned variable and the difference sum of the aforementioned images, weighted in a certain way, can be used as a loss function representing the gap.

[0125] Thirdly, as mentioned above, the differences in known image features can be applied. Classic image features include AKAZE (Accelerated KAZE), Cosine similarity, and histograms.

[0126] According to the determination process of the first embodiment, an image generation unit 132a, including a machine learning model (GAN), generates a pseudo-image with the characteristics of a normal product. Therefore, the determination unit 132 can obtain a pseudo-image of the molded product being inspected as normal with a small workload. Moreover, by calculating the difference between the input image of the molded product being inspected and the generated pseudo-image, the determination unit 132 can accurately determine the quality of the molded product.

[0127] Furthermore, according to the determination process of the first embodiment, a polarized image with four polarization channels is used as the input image and the pseudo image. Molded products from the injection molding machine 2 typically transmit a significant amount of light, and their quality can be distinguished based on the magnitude of their internal stress. Since the aforementioned polarized image reflects the internal stress of the light-transmitting molded product, by applying the aforementioned polarized image as the input image and the pseudo image, the quality determination of the molded products from the injection molding machine 2 can be achieved with high precision.

[0128] (Determination Unit of the Second Embodiment)

[0129] like Figure 10 As shown, the determination unit 132 includes: a processing unit 132p0, which calculates an input image from a polarization image; and an image generation unit 132a, which generates a pseudo image from the input image based on a machine learning model. The determination unit 132 also includes: a processing unit 132p1, which calculates the difference between the pseudo image and the input image; and a processing unit 132p2, which compares the difference with a threshold and determines the quality of the molded product 3 based on the input image, the calculated image, and the pseudo image.

[0130] The image generation unit 132a of the second embodiment inputs three images as input images: a monochrome image, a linear polarization degree image, and a linear polarization angle image of the molded article to be inspected. Furthermore, the image generation unit 132a generates the monochrome image, linear polarization degree image, and linear polarization angle image as pseudo images. Even if the molded article in the input image contains an anomaly, the image generation unit 132a will generate a pseudo image possessing the characteristics of a normal molded article. Specifically, the monochrome image, linear polarization degree image, and linear polarization angle image refer to a single image obtained by assigning the pixel values ​​of the monochrome image, the pixel values ​​of the polarization degree, and the pixel values ​​of the polarization angle to the three channels of each pixel.

[0131] The monochrome image, the linear polarization degree image, and the linear polarization angle image are equivalent to processed images obtained from a polarization image with four polarization channels. The monochrome image is an image where the pixel values ​​are the average of the values ​​of the same pixel in the four polarization channels. The linear polarization degree image is an image where the value of each pixel corresponds to the degree of linear polarization (DoLP) of that pixel. The degree of linear polarization (DoLP) can be calculated based on the values ​​of the four polarization channels. The linear polarization angle image is an image where the value of each pixel corresponds to the angle of linear polarization (AoLP) of that pixel. The angle of linear polarization (AoLP) can be calculated based on the values ​​of the four polarization channels. The generation of the processed image can be performed by the determination unit 132 (processing unit 132p0) or by the imaging device 110.

[0132] The image generation unit 132a includes a GAN as a learning model. This learning model is pre-loaded with images of multiple molded sample products without anomalies (i.e., normal) as training data, and machine learning is performed using these images. The multiple molded sample products may include a number of sample products with different states within the normal range. Through this machine learning, the image generation unit 132a generates pseudo-images possessing the characteristics of normal molded products.

[0133] If a pseudo-image is obtained, the determination unit 132 (processing unit 132p1) calculates the difference between the input image and the pseudo-image. Furthermore, the determination unit 132 (processing unit 132p2) determines that if the difference is below a threshold, the molded product is normal; if the difference exceeds the threshold, the molded product is abnormal.

[0134] In the output processing of the determination result (S507), the output unit 133 can output the input image, the pseudo image, and the determination result of whether it is normal or abnormal to the display unit 52. Moreover, the determination unit 132 can also display the difference between the input image and the pseudo image, as well as the areas with larger differences for each pixel.

[0135] Regarding the difference between the input image and the pseudo image, any quantity that can represent the difference can be used. For example, the three quantities mentioned above can be used. However, among the three quantities mentioned above, the first quantity representing the difference is changed to the sum of the differences between the input image and the pseudo image when the components of the monochrome image, the linear polarization degree image, and the linear polarization angle image are set as three channel components.

[0136] According to the determination process of the second embodiment, an image generation unit 132a, including a machine learning model (GAN), generates a pseudo-image with the characteristics of a normal product. Therefore, the determination unit 132 can obtain a pseudo-image of the molded product being inspected as normal with a small workload. Moreover, by calculating the difference between the input image of the molded product being inspected and the generated pseudo-image, the determination unit 132 can accurately determine the quality of the molded product.

[0137] Furthermore, according to the determination process of the second embodiment, a set of three images—a monochromatic image, a linear polarization degree image, and a linear polarization angle image—calculated from a polarization image with four polarization channels, are used as input images and pseudo-images. Molded products from injection molding machine 2 typically transmit a significant amount of light, and their quality can be distinguished based on the magnitude of internal stress. Since the aforementioned monochromatic image, linear polarization degree image, and linear polarization angle image reflect the internal stress of the light-transmitting molded product, by applying these images as input images and pseudo-images, the quality determination of molded products from injection molding machine 2 can be achieved with high precision.

[0138] (Determination Unit of the Third Embodiment)

[0139] like Figure 11 As shown, the determination unit 132 includes: a processing unit 132p0, which calculates an input image from a polarization image; and an image generation unit 132a, which generates a pseudo image from the input image based on a machine learning model. The determination unit 132 also includes: a processing unit 132p1, which calculates the difference between the pseudo image and the input image; and a processing unit 132p2, which compares the difference with a threshold and determines the quality of the molded product 3 based on the input image, the calculated image, and the pseudo image.

[0140] The image generation unit 132a of the third embodiment inputs a principal stress surface related image of the molded article to be inspected as an input image. Furthermore, the image generation unit 132a generates a pseudo-image from the stress-related image. Even if the molded article in the input image contains anomalies, the image generation unit 132a generates a pseudo-image possessing the characteristics of a normal molded article.

[0141] Principal stress surface related images refer to images such as phase difference images and principal stress difference images that are related to the principal stress surfaces of the molded product's internal stress. A phase difference image is an image where the pixel values ​​represent the phase difference. Here, the phase difference refers to the phase difference between polarized light perpendicular to the principal stress surface and polarized light parallel to the principal stress surface. A principal stress difference image is an image where the pixel values ​​represent the principal stress difference. The principal stress surface related images (phase difference images and principal stress difference images) are equivalent to processed images obtained from polarization images with four polarization channels. The generation of the processed images can be performed by the determination unit 132 (processing unit 132p0) or by the imaging device 110.

[0142] In addition, as a principal stress surface related image, an image that further includes the aforementioned monochrome image as a component of other channels can also be applied.

[0143] The image generation unit 132a includes a GAN as a learning model. This learning model is pre-loaded with images of multiple molded sample products without anomalies (i.e., normal) as training data, and machine learning is performed using these images. The multiple molded sample products may include a number of sample products with different states within the normal range. Through this machine learning, the image generation unit 132a generates pseudo-images possessing the characteristics of normal molded products.

[0144] If a pseudo-image is obtained, the determination unit 132 (processing unit 132p1) calculates the difference between the input image and the pseudo-image. Furthermore, the determination unit 132 (processing unit 132p2) determines that if the difference is below a threshold, the molded product is normal; if the difference exceeds the threshold, the molded product is abnormal.

[0145] In the output processing of the determination result (S507), the output unit 133 can output the input image, the pseudo image, and the determination result of whether it is normal or abnormal to the display unit 52. Moreover, the determination unit 132 can also display the difference between the input image and the pseudo image, as well as the areas with larger differences for each pixel.

[0146] Regarding the difference between the input image and the pseudo image, any quantity that can represent the difference can be used. For example, the three quantities mentioned above can be used. However, among the three quantities mentioned above, the first quantity representing the difference is changed to the sum of differences calculated for the values ​​of each pixel in the principal stress surface related image.

[0147] According to the determination process of the third embodiment, the image generation unit 132a, which includes a machine learning model (GAN), generates a pseudo-image with the characteristics of a normal product. Therefore, the determination unit 132 can obtain a pseudo-image of the molded product being inspected as normal with a relatively small workload. Moreover, by calculating the difference between the input image of the molded product being inspected and the generated pseudo-image, the determination unit 132 can accurately determine the quality of the molded product.

[0148] Furthermore, according to the determination process of the third embodiment, the principal stress surface correlation image is used as the input image and pseudo image. Molded products from the injection molding machine 2 typically transmit a large amount of light, and their quality is distinguished based on the magnitude of internal stress. Since the aforementioned principal stress surface correlation image reflects the internal stress of the light-transmitting molded product, by applying the aforementioned principal stress surface correlation image as the input image and pseudo image, the quality determination of the molded products from the injection molding machine 2 can be achieved with high precision.

[0149] (Determination Unit of the Fourth Embodiment)

[0150] like Figure 12 As shown, the determination unit 132 includes: a processing unit 132p0, which calculates an input image from a polarization image; and an image generation unit 132a, which generates a pseudo image from the input image based on a machine learning model. The determination unit 132 also includes: a processing unit 132p1, which calculates the difference between the pseudo image and the input image; and a processing unit 132p2, which compares the difference with a threshold and determines the quality of the molded product 3 based on the input image, the calculated image, and the pseudo image.

[0151] The image generation unit 132a of the fourth embodiment inputs a relative stripe level image of the molded article to be inspected as an input image. Furthermore, the image generation unit 132a generates a relative stripe level image as a pseudo image. Even if the molded article in the input image contains an anomaly, the image generation unit 132a will generate a pseudo image possessing the characteristics of a normal molded article.

[0152] A relative fringe order image refers to an image of photoelastic fringes obtained through a polarizer, which is equivalent to a processed image obtained from a polarization image with four polarization channels. The generation of the processed image can be performed by the determination unit 132 (processing unit 132p0) or by the imaging device 110.

[0153] The image generation unit 132a includes a GAN as a learning model. This learning model is pre-loaded with images of multiple molded sample products without anomalies (i.e., normal) as training data, and machine learning is performed using these images. The multiple molded sample products may include a number of sample products with different states within the normal range. Through this machine learning, the image generation unit 132a generates pseudo-images possessing the characteristics of normal molded products.

[0154] If a pseudo-image is obtained, the determination unit 132 (processing unit 132p1) calculates the difference between the input image and the pseudo-image. Furthermore, the determination unit 132 (processing unit 132p2) determines that if the difference is below a threshold, the molded product is normal; if the difference exceeds the threshold, the molded product is abnormal.

[0155] In the output processing of the determination result (S507), the output unit 133 can output the input image, the pseudo image, and the determination result of whether it is normal or abnormal to the display unit 52. Moreover, the determination unit 132 can also display the difference between the input image and the pseudo image, as well as the areas with larger differences for each pixel.

[0156] Regarding the difference between the input image and the pseudo-image, various quantities can be used as long as they can represent the difference. For example, the three quantities mentioned above can be used. However, among the three quantities mentioned above, the first quantity representing the difference is changed to the sum of differences calculated with respect to the values ​​of each pixel in the relative stripe level image.

[0157] According to the determination process of the fourth embodiment, an image generation unit 132a, including a machine learning model (GAN), generates a pseudo-image with the characteristics of a normal product. Therefore, the determination unit 132 can obtain a pseudo-image of the molded product being inspected as normal with a relatively small workload. Moreover, by calculating the difference between the input image of the molded product being inspected and the generated pseudo-image, the determination unit 132 can accurately determine the quality of the molded product.

[0158] Furthermore, according to the determination process of the fourth embodiment, a relative stripe level image is used as both the input image and the pseudo-image. Molded products from the injection molding machine 2 typically transmit a significant amount of light, and their quality is determined based on the magnitude of internal stress. Since the aforementioned relative stripe level image reflects the internal stress of the light-transmitting molded product, by applying the aforementioned relative stripe level image as both the input image and the pseudo-image, the quality determination of the molded products from the injection molding machine 2 can be achieved with high precision.

[0159] (Decision Unit of the Fifth Embodiment)

[0160] like Figure 13 As shown, the determination unit 132 includes: a processing unit 132p0 that calculates a relative stripe level image from a polarization image; a first image generation unit 132b and a second image generation unit 132c that generate images based on a machine learning model. The determination unit 132 further includes: a processing unit 132p1 that calculates the difference between the input image of the second image generation unit 132c and the pseudo image; and a processing unit 132p2 that compares the difference with a threshold and determines the quality of the molded product 3 based on the input image, the calculated image, and the pseudo image.

[0161] The first image generation unit 132b takes into input a relative stripe level image of the molded article to be inspected and outputs a stress distribution image that accurately reflects the stress distribution of the molded article. The first image generation unit 132b includes a Generative Adversarial Network (GAN) as a first learning model. Machine learning is performed in the first learning model by pre-providing relative stripe level images and stress distribution images of multiple sample molded articles as training data, and outputting the corresponding stress distribution image when a relative stripe level image is input. The multiple sample molded articles may include a number of sample molded articles with different states within a range encompassing both normal and abnormal conditions. Through this machine learning, the first image generation unit 132b is able to output a stress distribution image that accurately reflects the stress distribution of the molded article to be inspected.

[0162] The second image generation unit 132c inputs an image of the stress distribution of the molded part being inspected as an input image. Furthermore, even if the molded part being inspected contains anomalies, the image generation unit 132a generates a pseudo-image of the stress distribution that possesses the characteristics of a normal molded part.

[0163] The stress distribution image input to the second image generation unit 132c is an image generated by the first image generation unit 132b based on a relative stripe order image obtained from polarization images of the four polarization channels. Therefore, the stress distribution image input to the second image generation unit 132c is equivalent to a processed image obtained from polarization images of the four polarization channels.

[0164] The second image generation unit 132c includes a GAN as a second learning model. The second learning model is pre-loaded with images of multiple molded sample products without anomalies (i.e., normal) as training data, and performs machine learning using these images. The multiple molded sample products may include a number of sample products with different states within the normal range. Through this machine learning, the second image generation unit 132c generates pseudo-images possessing the characteristics of normal molded products.

[0165] If a pseudo-image is obtained, the determination unit 132 (processing unit 132p1) calculates the difference between the input image and the pseudo-image. Furthermore, the determination unit 132 (processing unit 132p2) determines that if the difference is below a threshold, the molded product is normal; if the difference exceeds the threshold, the molded product is abnormal.

[0166] Figure 14This is an image representing an example of the inspection result. In the output processing of the determination result (S507), the output unit 133 can output the input image pin input to the second image generation unit 132c, the pseudo image pout output by the second image generation unit 132c, and the determination result e1 indicating whether it is normal or abnormal to the display unit 52. Furthermore, the determination unit 132 can also display the difference e2 between the input image pin and the pseudo image pout, and the regions e3 where the difference is large for each pixel.

[0167] Regarding the difference between the input image and the pseudo image, any quantity that can represent the difference can be used. For example, the three quantities mentioned above can be used. However, among the three quantities mentioned above, the first quantity representing the difference is changed to the sum of differences calculated from the values ​​of each pixel in the stress distribution image.

[0168] According to the determination process of the fifth embodiment, a second image generation unit 132c, including a second learning model (GAN) with machine learning capabilities, generates a pseudo-image with the characteristics of a normal product. Therefore, the determination unit 132 can obtain a pseudo-image of the molded product being inspected as normal with a relatively small workload. Moreover, by calculating the difference between the input image of the molded product being inspected and the generated pseudo-image, the determination unit 132 can accurately determine the quality of the molded product.

[0169] Furthermore, according to the determination process of the fifth embodiment, a stress distribution image is used as both the input image and the pseudo-image. The quality of the molded articles from the injection molding machine 2 is determined based on the magnitude of the internal stress. Since the stress distribution image reflects the internal stress, by using it as both the input image and the pseudo-image, the quality determination of the molded articles from the injection molding machine 2 can be achieved with high precision.

[0170] Machine Learning Processing

[0171] Figure 15 This is a flowchart illustrating an example of the sequence of machine learning processes performed by the inspection device. Figure 16 (A) is a diagram illustrating the machine learning processing of the first to fourth embodiments. Figure 16 (B) is a diagram illustrating the machine learning process of the fifth embodiment.

[0172] The aforementioned image generation unit 132a, first image generation unit 132b and second image generation unit 132c need to be pre-loaded with training data to perform machine learning.

[0173] Therefore, as Figure 2As shown, the inspection device 100 includes: an operation unit 141 capable of selecting an operation mode (machine learning mode) for performing machine learning; and a training data input unit 142 capable of inputting training data from an external source. Furthermore, the inspection device 100 includes: a CAE calculation unit 143 for generating training data; and a molding step data input unit 144 capable of inputting molding step data.

[0174] like Figure 15 As shown, the operator selects the machine learning mode via the operation unit 141 and can enter the machine learning process (S601). If the machine learning process is entered, the operator inputs training data from the training data input unit 142 (S602). Alternatively, the operator can input molding step data from the molding step data input unit 144 (S603). Then, based on the molding step data, the CAE calculation unit 143 generates training data (S604). Furthermore, the determination unit 132 receives the training data and causes the image generation unit 132a (or the first image generation unit 132b and the second image generation unit 132c) to perform machine learning (S605).

[0175] Machine learning can be performed each time the shape, material, or both of the molded part produced by injection molding machine 2 changes. A change in the shape of the molded part refers to a change in the mold.

[0176] The images used as training data can be polarized images obtained by actually photographing the sample molded article with the camera device 110, or computational images obtained by performing calculations (including analysis) on the polarized images. The sample molded article can be a molded article with the same shape and material as the molded article 3 that is the object of inspection.

[0177] Furthermore, the image used as training data can be a simulated image of the molded sample (virtual molded product) obtained by CAE (Computer-Aided Engineering) calculations based on the molding step data of injection molding machine 2, and a virtual computational image representing the stress distribution of the simulated molded sample obtained by calculation based on these characteristics. The molding step data is data capable of simulating the molding steps, and includes at least data on the shape of the mold cavity.

[0178] exist Figure 2 In the example shown, the inspection device 100 is equipped with a CAE calculation unit 143 that performs the above-mentioned CAE calculation and generates images of training data within the inspection device 100. However, the CAE calculation unit 143 may also be configured as a computer separate from the inspection device 100.

[0179] In the first embodiment, polarization images of four polarization channels from multiple molded sample articles are used as training data. These multiple molded sample articles can be normal articles and differ in state within a normal range.

[0180] In the second embodiment, as training data, images consisting of monochrome images, linear polarization degree images, and linear polarization angle images of multiple molded sample articles are used as a group. These multiple molded sample articles can be normal products, but their states differ within a normal range.

[0181] In the third embodiment, principal stress surface correlation images (phase difference images, principal stress difference images, etc.) of multiple sample molded articles are used as training data. These multiple sample molded articles can be normal articles and differ in state within a normal range.

[0182] In the fourth embodiment, relative stripe order images of multiple sample molded articles are applied as training data. These multiple sample molded articles can be normal articles, but their states differ within a normal range.

[0183] In the fifth embodiment, as training data for the first image generation unit 132b, a combination of the relative stripe level image and the stress distribution image of each of the multiple molded sample articles is applied. The first image generation unit 132b performs machine learning by inputting the relative stripe level image and outputting the corresponding stress distribution image. The aforementioned multiple molded sample articles may include normal articles and abnormal articles, and their states are different.

[0184] As training data for the second image generation unit 132c, stress distribution images of multiple molded sample articles are applied. These multiple molded sample articles can be normal articles, but their states differ within the normal range.

[0185] Through the aforementioned machine learning, the determination unit 132 can perform the aforementioned determination process.

[0186] As described above, according to the inspection apparatus 100 and injection molding system 1 of this embodiment, the imaging device 110 acquires a polarization image of the molded article 3, which is the object of inspection. Furthermore, the image generation unit 132a or the second image generation unit 132c of the determination unit 132 inputs the polarization image or a processed image obtained from the polarization image through calculation as an input image, and outputs a pseudo-image having the characteristics of a normal article. Moreover, the determination unit 132 determines the quality of the molded article based on the input image and the pseudo-image. Therefore, the determination unit 132 can perform high-precision quality determination with a relatively small workload.

[0187] Furthermore, the inspection device 100 has a training data input unit 142 for inputting training data, and inputs the following images as training data: a polarization image of a sample molded article that is at least a normal product, a computed image obtained from the polarization image, a virtual polarization image calculated on a virtual molded article calculated based on molding step data and simulated by CAE calculation, and a computed image obtained from the virtual polarization image. Moreover, the image generation unit 132a (or the second image generation unit 132c) uses this training data to perform machine learning. Through such machine learning, the aforementioned pseudo-image generation can be achieved.

[0188] Furthermore, the polarization image or virtual polarization image includes four or more polarization channels with different polarization angles. Additionally, the processed image includes one or more of the following: a monochrome image, a combination of a linear polarization degree image and a linear polarization angle image, a principal stress surface correlation image, a relative fringe order image, and a stress distribution image. By using such images, the quality of the molded product from the injection molding machine 2 can be determined with high precision regarding transmitted light.

[0189] The embodiments of the present invention have been described above. However, the present invention is not limited to the above embodiments. The detailed structure shown in the embodiments can be appropriately modified without departing from the spirit of the invention.

Claims

1. An inspection device comprising: The camera device acquires polarization images of multiple polarization channels of the molded article; and The determination unit determines the quality of the molded article. The determination unit includes an image generation unit that generates pseudo-images from input images based on a machine learning model. The determination unit inputs the polarization image or a processed image obtained from the polarization image as the input image to the image generation unit, and determines the quality of the molded product based on the input image and the pseudo image generated by the image generation unit. The image generation unit generates a pseudo image with the same number of polarization channels based on the polarization images or the processed images of the multiple polarization channels. The determination unit compares the polarization image or the processed image with the pseudo image in the same polarization channel, and determines whether it is good or bad based on the comparison result.

2. The inspection device according to claim 1, comprising: The training data input section is for inputting training data for machine learning. The training data includes at least one of the following: a polarization image of a normal molded product, a computed image obtained from the polarization image of the normal molded product, a virtual polarization image of a virtual molded product obtained by performing calculations based on molding step data that can simulate normal molding steps, and a computed image obtained from the virtual polarization image. The learning model uses the input training data to perform machine learning.

3. The inspection device according to claim 2, wherein, The polarization image or the virtual polarization image includes four or more polarization channels with different polarization angles.

4. The inspection device according to claim 2, wherein, The computational image includes one or more of the following: a monochrome image, a combination of a linear polarization degree image and a linear polarization angle image, a principal stress surface correlation image, a relative fringe series image, and a stress distribution image.

5. An injection molding system comprising: Injection molding machine; and The inspection device according to any one of claims 1 to 4 is used to inspect molded articles formed by the injection molding machine.

6. An inspection method, wherein, Polarization images of multiple polarization channels of the molded product are acquired using a camera device. The polarization image or the processed image obtained from the polarization image is input to the image generation unit as an input image. The quality of the molded product is determined based on the pseudo-image of the input image generated by the image generation unit and the input image itself. The image generation unit generates a pseudo image with the same number of polarization channels based on the polarization images or the processed images of the multiple polarization channels. The polarization image or the processed image is compared with the pseudo image in the same polarization channel, and the quality is determined based on the comparison result.

Images